Survive the AI Knife Fight: Building Products That Win — Brian Balfour, Reforge

Key Concepts

• AI Product Strategy
• Competitive Advantage in AI
• Data as a Differentiator
• Functionality as a Differentiator
• Unmet Customer Needs
• AI Lego Blocks (Composable AI)
• Sequencing Moats
• Granola as a Case Study

Main Topics and Key Points

The Hyper-Competitive Landscape

• Rapid Product Launches: The tech industry is experiencing an unprecedented rate of product launches and feature releases, particularly in AI. Examples include Notion, Figma, Atlassian, Google, and OpenAI launching competitors in various spaces.
• Company Collapse: Companies are collapsing at an accelerated rate (months instead of years), exemplified by Chegg and Stack Overflow's decline after ChatGPT's launch.
• The Core Question: The fundamental question that product teams must answer is: "What do I build, and why will it win?" This question has become significantly harder to answer due to the intense competition.
• Competitive Map: The current competitive environment is likened to the Battle of Gettysburg, with fast-moving incumbents (Microsoft, Google, Meta), horizontal platforms (ChatGPT, Anthropic), foundational technology shifts, and numerous startups vying for market share.

The Role of Product and Competitive Advantage

• Product's Core Job: The primary role of product is to identify and address unmet customer needs, a role that has been obscured by project management and agile processes.
• Sean Claus Quote: "You're constantly trying to get ahead, you're trying to find the angle, the question that has not yet been asked that gives you an insight that is not being actioned by other people... that is your competitive advantage."
• Competitive Advantage: Competitive advantage comes from insights that others are not acting on.

Avoiding Common Traps

• Trap 1: Reinventing the AI Wheel: Building custom AI models and infrastructure is often unnecessary.
• Trap 2: Copying Basic AI Features: Simply implementing basic AI features like chatbots is insufficient.
• The Solution: Composable AI: Treat AI as a series of Lego blocks, assembling differentiated AI features and products by integrating the best available AI capabilities with your product's data and functionality.

Anatomy of a Winning AI Product

• Three Key Building Blocks:
  • AI Capabilities: Pre-trained AI models, audio processing, image processing, etc. These are generally accessible to everyone and do not provide differentiation.
  • Data: Provides context to the AI model to generate unique outputs. Types of data include real-time, user-specific, domain-specific, human judgment (curation), and reinforcement data. The value lies in the marginal value of your data over what is already trained in the models.
  • Functionality: Determines how the AI behaves and gives your AI product superpowers. Includes specialized workflows, unique algorithms, business rules, and integrations.
• Systemic Connection: The three building blocks must work together as a system. Data informs the AI's understanding, leading to unique outputs, which in turn generate more unique data. Functionality controls the AI's actions and interacts with the AI to create a delightful user experience. AI can also call tools within the product's functionality.

Granola Case Study

• Market Entry: Granola entered a crowded AI note-taking market with incumbents like Fathom, Otter, and Fireflies.
• Unmet Customer Need: Granola focused on empowering users to take better notes, rather than replacing the note-taking process entirely.
• AI Lego Blocks Assembly:
  • Data: Granola's data includes user-taken notes and transcriptions.
  • AI Capabilities: They used off-the-shelf capabilities like Deepgram for transcription and Anthropic/OpenAI for other functionalities.
  • Functionality: A Mac app that detects meeting starts, accesses system sound for transcription, and integrates with calendars for metadata.
• Flywheel Effect: The unique output (enhanced notes) creates a repository of data that enables additional features like chatting across meetings and project workspaces.
• Sequencing: Granola is continuously sequencing new features and integrations (project workspaces, CRM integration, company wiki) to maintain a competitive edge.

Sequencing Moats

• Jamon Ball Quote: "The real moat is just a sequence of smaller moats stacked together. Each one buys time. What you do with that time, how fast you execute, how quickly you evolve determines whether you stay ahead. If the moat used to be 6 to 12 months, today it's 2 to 3 weeks."

Conclusion

• Recap: To win in AI, answer these questions:
  • What are your unmet customer problems?
  • What AI capabilities can solve those problems in novel ways?
  • What proprietary data can power those solutions?
  • What superpowers can your product give to AI?
• Assemble these three foundational Lego blocks to create a differentiated AI product.